MANY MINDEDChatGPT's free tier now defaults to simplified model
OpenAI began rolling out ChatGPT updates on August 7, 2026, changing the free tier default to GPT-5.6 Luna—a smaller model in the GPT-5.6 family. This model shows 62% fewer factual errors in internal tests compared to the previous free tier (GPT-5.5 Instant), though it delivers less complex responses than the paid GPT-5.6 Sol variant. The update explicitly states Luna is a 'step up' from GPT-5.5 Instant but not a weaker experience than prior free tiers. Free users gain unlimited text chats and a 'Think' button for complex queries next week, while file uploads and image generation remain unchanged.
This shift creates friction in knowledge access: the simplified model prioritizes speed and basic accuracy for casual users but restricts advanced reasoning and nuanced outputs. For those needing deeper analysis, the paid Sol model—designed for complex tasks with better factual precision—remains gated. OpenAI clarifies Luna does not provide access to frontier reasoning models, meaning users seeking high-stakes problem-solving face a trade-off between simplicity and capability.
The source provides no detail on how this affects real-world knowledge costs or adoption rates. The internal error metrics are unverified, and Sol remains unavailable in ChatGPT Work or Codex.
Source: The Decoder